64 AvailableNew on-lineZealand at: Journal http://www.newzealandecology.org/nzje/ of Ecology, Vol. 36, No. 1, 2012

Breeding variation in female (Strigops habroptilus) on Codfish Island in a year of low food supply

Joanna Whitehead1*, Brad Case, Kerry-Jayne Wilson2 and Laura Molles Department of Ecology, Faculty of Agriculture and Life Sciences, Lincoln University, PO Box 84, Lincoln, 1Present address: Department of Conservation, PO Box 29, Te Anau 9600, New Zealand 2Present address: PO Box 70, Charleston 7865, West Coast, New Zealand *Author for correspondence (Email: [email protected])

Published on-line: 30 August 2011

Abstract: We investigated why some mature females of New Zealand’s critically endangered parrot, the kakapo (Strigops habroptilus), did not attempt to breed during the 2005 breeding season on Codfish Island. At a population level, the initiation of kakapo breeding appears to correspond with years of mast fruiting of rimu (Dacrydium cupressinum) trees, with the proportion of females that breed each season dependent on the quantity of rimu fruit available. This research investigates possible links between habitat quality within individual home ranges and the breeding status of adult females during 2005, when the abundance of available rimu fruit was low. We assessed the importance of both home range size and habitat characteristics in determining breeding attempts. Foraging home ranges were characterised using radio-tracking and triangulation techniques. The relative importance of habitat variables in optimal breeding habitat was assessed using ecological niche factor analysis. Our results show that female kakapo breeding in 2005 had, on average, home ranges twice the size of those females that did not breed that season and the ranges contained a significantly greater quantity of mature rimu forest. Multivariate analysis illustrates female kakapo were effectively partitioning available habitat, as breeders’ foraging locations were positively correlated with high-abundance rimu forest with a tall canopy, described as optimal breeding habitat. In contrast non-breeders’ locations were weakly correlated with short forest containing little or no mature rimu forest. To maximise reproductive output each breeding season, conservation managers need to ensure that all breeding-aged females occupy optimal breeding habitat on Codfish Island. Removal to other islands of kakapo not required in the breeding population may enable females to increase their home range size and occupy better breeding habitat. Keywords: ecological niche factor analysis; habitat selection; habitat quality; home range size; radio tracking; reproduction; rimu

Introduction either the patchiness of the fruit crops that trigger breeding or differences in female condition (Eason et al. 2006). This Despite having an annual gonadal cycle (Cockrem & Rounce research investigates the hypothesis that variation in breeding 1995; Cockrem 2006), kākāpō (Strigops habroptilus) breed attempts between females in low fruit years is associated with only once every 2–5 years on Codfish Island, the location of the the quality and quantity of critical food resources available entire population of breeding-aged females, in response to the within their individual foraging ranges. mast fruiting of rimu (Dacrydium cupressinum) trees (Elliott To test this hypothesis we estimate foraging home ranges et al. 2001; Eason et al. 2006). The recovery of this critically of adult female kākāpō on Codfish Island in a non-breeding endangered parrot is slowed even further by considerable year (2006) and compare home range characteristics between variation in the number of females that nest in some rimu- females that did and did not breed in the previous (2005) fruiting years. On a population level, breeding in this lek summer. We predict that individual female kākāpō that bred will species (Merton et al. 1984) appears to be correlated with the have larger foraging home ranges, since having a larger home quantity of rimu fruit available, with most females nesting in range should increase the likelihood that an individual will be years of high rimu fruit production (Elliot et al. 2006; Table 1). able to access more abundant, higher quality resources required However, in low rimu fruiting years only some female kākāpō for breeding (Millspaugh & Marzluff 2001). We also predict in the population nest (Elliot et al. 2006), causing concern for that there will be distinct differences in vegetation composition conservation managers as there is no explanation for what and environmental conditions within the home ranges of might be preventing some females from breeding. breeding compared with non-breeding individuals, reflecting Hypotheses advanced to explain what triggers kākāpō to differences in habitat quality and resource availability. breed all focus on the females receiving cues that there will Specifically we investigate whether (1) foraging home be sufficient ripe rimu fruit to raise their young. Whether range sizes or vegetation selection differ between females that these triggers are cognitive (Lignon 1974) or nutritional bred the previous summer and those that did not, (2) tenure (Powlesland et al. 1992), or mediated via weight gain (Harper on the island correlates with home range size or nesting, et al. 2006) or hormonal stimulation (Cockrem 2006; Fidler (3) the prevalence of vegetation types with a high rimu et al. 2008) remains unresolved. In a given breeding year, component inside foraging ranges differs between breeding variation in the proportion of females that breed may reflect and non-breeding females, and (4) the relative importance of

New Zealand Journal of Ecology (2012) 36(1): 64-74 © New Zealand Ecological Society. Whitehead et al.: Breeding variation in female kakapo 65

Table 1. Breeding years on Codfish Island (since the first a range of habitat variables in characterising the ecological translocation of kākāpō in 1987) showing the level of the niche required for females to breed in a low-rimu-mast year. rimu mast, percentage of rimu branches bearing fruit and The results of these analyses assist us in predicting optimal the number of breeding-aged females that nested (Kakapo breeding habitat for female kākāpō on Codfish Island, thus Recovery Programme, unpubl. data) compared with the providing an additional tool for managing this critically total number on the island. The breeding year of 1992 is endangered species. not included as rimu mast and breeding attempts were not accurately recorded. ______Breeding Rimu Rimu bearing No. of breeding-aged Methods year mast fruit females* that nested level (%) (total no. on island) Codfish Island / Whenuahou (1475 ha) is situated 3 km off the ______north-west coast of Stewart Island in southern New Zealand 1997 Low 14 6 (10) (46°46.07′ S, 167°37.23′ E; Fig. 1). Habitat on the island is 2002 High 35 20 (21) dominated by mixed-podocarp forest with pakahi scrub in 2005 Low 11 10 (21) elevated areas and dense scrub around coastal margins (Meurk & Wilson 1989). At the time of this research in 2006, Codfish 2008 Low 13 5 (38) Island was home to the only known breeding population of 2009 High 39 27 (38) ______kākāpō as all 21 adult females of breeding age (9 years or *Note breeding age was defined as 9 years or older prior to 2008 older; Eason et al. 2006) resided there, along with 20 adult (Eason et al. 2006), and as 6 years or older from the 2008 breeding males and 13 juveniles. Our research involved 18 of the 21 season onwards as 6-year-old females nested that year (Moorhouse breeding-aged females; one female was excluded from the 2009). If a breeding age of 6 years or older had been used for the study because she was hand-reared and had formed attachments earlier years there would have been 26 breeding-aged females in to staff and two others were excluded because they were in 2005, but no change to the 2002 or 1997 figures. inaccessible locations on the island. We assigned a breeding status (i.e. breeder or non-breeder) to each of the remaining 18 adult females, based on the ’ breeding status from the previous year (2005), which was the most recent breeding year (Table 1).

Stewart Island

Figure 1. Codfish Island / Whenuahou, located 3 km off the north-west coast of Stewart Island, New Zealand (NZMS260 1:50 000 scale map series provided by Land Information New Zealand). 66 New Zealand Journal of Ecology, Vol. 36, No. 1, 2012

Radio tracking of foraging locations 1989; Harris et al. 1990). We used a fixed kernel and the least- All kākāpō on the island are fitted with back-mounted radio squares cross-validation technique (Seaman & Powell 1996). transmitters, for management purposes. Foraging locations Home ranges were estimated using the software Ranges6 were estimated for each of the 18 female kākāpō between v.1.2 (Kenward et al. 2003). We used a regression model to 15 March and 30 May 2006, using standard radio-tracking determine whether the number of locations used to estimate techniques (Clout 2006). We used TR4 radio receivers a foraging range influenced its size. (Teleonics, AZ, USA) and three-element Yagi hand-held aerials We analysed the influence of 2006 foraging home range fitted with sighting compasses (Sirtrack, Havelock North, NZ) size on breeding status, which was determined using breeding to locate females at night. Once located, a female’s foraging information collected from the summer of the previous year position was estimated from the walking tracks on the island, (2005). Previous analyses (Whitehead 2007) had shown using triangulation (White & Garrott 1990). Triangulations there to be no significant difference between home ranges of were estimated by recording the direction of the transmitters’ females in breeding and non-breeding years for these birds strongest signal as a bearing from marked locations along on Codfish Island. This finding is consistent with previous the tracks and intersecting these bearings using an existing research showing that kākāpō generally stay within similar Microsoft Access database designed by the Department of home ranges for a number of years (Merton et al. 1984; Conservation’s Kakapo Recovery Programme (DOC unpubl.). Moorhouse & Powlesland 1991; Powlesland et al. 1992) and Up to 10 bearings were recorded in the field with the five having that breeding is not thought to alter home range size (Farrimond the closest intersection used for the final location estimate. et al. 2006). We were unable to make comparisons with 2007 As triangulations are location estimates rather than exact as no breeding occurred that year (Table 1). As a further test fixes (Springer 1979) we used a number of approaches to of this assertion, we plotted nesting locations from 2005 onto minimise errors and obtain the greatest possible data accuracy the 2006 home range locations for each female. Each female’s (Whitehead 2007). To ensure we were estimating foraging nest fell inside (or within 100 m of) 95% kernel foraging rather than roosting locations we collected data during the ranges, supporting the idea that kākāpō use similar home night but not within 2 h of sunset or sunrise. The time of night ranges between breeding and non-breeding years. Statistical that individuals were tracked was varied throughout the study comparisons were made between home range sizes between period. Bearings were taken from as close as possible and breeding and non-breeding female kākāpō. Two-sample t-tests from a variety of angles around the while the researcher were used to compare mean home range sizes for MCP and remained on formed tracks. 95% kernels. The 75% and 50% kernels were not normally Bearings were recorded simultaneously by two or more distributed so were compared using the non-parametric people situated at different locations along the tracks, or by Wilcoxon rank sum test (unpaired). one person moving quickly between locations. It is unlikely Initially, we considered whether other factors such as age, that a kākāpō would have moved any significant distance body condition or time spent on the island (tenure) may have during the time taken to complete a triangulation as bearings influenced a female’s home range size or breeding ability, but were not recorded if the transmitter signal indicated the bird due to insufficient data we were only able to test the effect was moving (Mech 1983). If a kākāpō did move a significant of tenure. Using a Wilcoxon rank sum test (with a continuity distance, the triangulation would not have been accurate enough correction) we tested whether tenure was correlated with her to include in later analysis, as any triangulations that had an breeding status during the 2005 breeding season. A Kendall’s estimated location triangle greater than 40 m in width were rank correlation was used to determine whether tenure was excluded. If movement did occur, we expect that the location correlated with foraging home range sizes. error would have been relatively low since kākāpō are known to move at rates of up to 50 m h–1 while foraging (Walsh Vegetation use analysis 2002) and the mean time taken to complete a triangulation The vegetation of Codfish Island was mapped in 2005 using was 14.6 ± 7.6 min. a combination of aerial photos, infrared images and ground We calculated the error of our triangulation technique by surveys (Kakapo Recovery Programme, DOC unpubl. data). comparing the estimated and actual locations of 25 transmitters Sixteen vegetation types were described according to the main hidden in the forest at similar locations and distances from canopy and subcanopy species, with some reference to the where we encountered kākāpō (Walsh et al. 2006). Occasional understorey vegetation (Table 2). To enable statistical analysis sightings of kākāpō on or near tracks were recorded using of vegetation selection by kākāpō, the original 16 vegetation a handheld global positioning unit (GPS) and recorded as types were combined into five aggregate classes based on the additional locations. Only one foraging location was collected similarity of the species and area of the island occupied by each per bird per night to allow for independence of the data. vegetation type. All vegetation types occupying 10% or less of the island’s area were included as an ‘other’ class, along with Estimating foraging home ranges the four vegetation types not used by adult female kākāpō in this We estimated foraging home ranges using two estimation study. A chi-squared test was used to determine if vegetation techniques that had previously been used in kākāpō home range selection by female kākāpō (within 75% kernel home ranges) studies (Trinder 1998; Farrimond et al. 2006, Walsh 2006) differed from the available vegetation on the island. – minimum convex polygons (MCP) and kernels (50%, 75% To investigate the importance of mature rimu trees for and 95%). The MCP method connects the outermost locations breeding in individuals, a second map illustrating the relative to form a polygon and is one of the most widely used methods abundance of mature rimu trees was created by combining of estimating home range (Harris et al. 1990; Kernohan et al. vegetation types described to have a high, moderate or low 2001). Kernel estimators are being increasingly used in home abundance of mature rimu trees (Table 2, Fig. 3). The areas range analysis as they provide information on how different of low-, medium- or high-abundance rimu forest in the parts of the home range are used disproportionately (Worton foraging ranges (75% kernel) of breeders and non-breeders Whitehead et al.: Breeding variation in female kakapo 67

Table 2. The original 16 vegetation types on Codfish Island reclassified into three columns for analysis: aggregated classes; rimu abundance and canopy height (Kakapo Recovery Programme, New Zealand Department of Conservation unpubl. data). ______Original vegetation Description Aggregated Rimu Canopy types classes abundance height (m) ______Miro–rimu Dense miro (Prumnopitys ferruginea) and rimu forest, > 20 m Rimu–miro High 20 tall, with a predominance of miro. Additional secondary species include kāmahi (Weinmannia racemosa), rātā (Metrosideros umbellata) and occasional tōtara (Podocarpus hallii). Rimu–miro Dense rimu forest, > 20 m tall. Secondary species include miro High 20 and rātā, but kāmahi can be locally common. Interspersed with occasional tōtara. ______Rātā Predominantly rātā forest, typically < 5 m tall, often with Rātā– Low 5 patches of mānuka (Leptospermum scoparium). Understorey podocarp often consists of Dracophyllum spp. Rātā–podocarp Short rātā-dominated forest interspersed with podocarps that Moderate 5 short are generally < 5 m tall. Occasional kāmahi. Understorey is commonly Dracophyllum spp. Possibly regenerating forest. ______Mixed podocarp Predominantly mixed rimu, miro and tōtara forest, 10–20 m Mixed High 15 stunted tall, with numerous rātā, occasional kāmahi and an understorey podocarp often consisting of Dracophyllum spp. stunted ______Coastal daisy Daisy forest scrub with Olearia spp. and Dracophyllum spp. Coastal Low 5 daisy – pakahi scrub Pakahi scrub Mānuka and Dracophyllum spp. scrub, predominantly 1–2 Low 5 m tall, interspersed with rātā, Olearia spp. and mingimingi (Cyathodes juniperina). Mostly in pakahi soils. ______Coastal scrub Scrub with strong coastal influence including Brachyglottis Other Low 5 spp., broadleaf (Griselinia littoralis), Hebe spp. and kāmahi. In wetter areas fern. Rimu–rātā Rimu forest interspersed with rātā and miro. Canopy height High 20 is typically > 20 m. Rātā dominates over miro as the predominant secondary species. Kāmahi–podocarp Mixed kāmahi–podocarp forest typically > 20 m tall, Moderate 20 with occasional rātā. Kāmahi is a canopy species and comprises of approximately half the forest-type composition. Rātā–podocarp Tall rātā-dominated forest interspersed with podocarps that are Moderate 15 generally < 10 m tall. Occasional kāmahi. Understorey is commonly Dracophyllum spp. Mixed podocarp tall Predominantly mixed rimu, miro and tōtara forest generally High 20 > 20 m, with some rātā. Found in the valley floor. Typically no Dracophyllum spp. Widespread podocarp seedlings. Kāmahi* Predominantly kāmahi forest, often in pure stands, but Moderate 15 occasionally interspersed with podocarps and rātā. Kāmahi–rātā* Predominantly kāmahi forest with frequent rātā. Also Moderate 15 occasionally interspersed with podocarps. Mānuka–broadleaf* Mix of mānuka, broadleaf, and Hebe found around the hut. Low 5 Sand dune* Sand dunes Low 5 ______*indicates vegetation types not occupied by adult female kākāpō in this study.

were compared using a non-parametric Wilcoxon rank sum Factor Analysis (ENFA; Hirzel et al. 2002) implemented in test (unpaired). the software package Biomapper 3.2 (Hirzel et al. 2006a). All statistical tests described above were carried out ENFA predicts the ecological niche that a species occupies using R statistical software v. 2.9 (R Development Core by contrasting the average value of a habitat variable across Team 2008). the study area with the average value in the cells occupied by the species, with any difference in these two values indicating Multivariate habitat analysis habitat selection. As habitat variables are not independent, a The relative importance of a range of habitat variables in factor analysis is used in ENFA to initially transform correlated predicting optimal breeding habitat for adult female kākāpō variables into the same number of uncorrelated factors (Brotons on Codfish Island was assessed using Ecological Niche et al. 2004), allowing the overall information explaining 68 New Zealand Journal of Ecology, Vol. 36, No. 1, 2012

the ecological niche of the species to be represented by two The quality of the ENFA models was assessed by uncorrelated indices: marginality and specialisation. determining how they differed from a random model of kākāpō The marginality value for each habitat variable is the distribution relative to available habitat. Model accuracy ecological distance between the species optimum and the was assessed using the continuous Boyce Index (Hirzel et al. mean within the study area (Hirzel et al. 2002). The larger 2006b), a 10-fold cross-validation procedure that spatially the absolute value of marginality, the more the species mean partitions the species dataset into independent partitions (Manly differs from the mean in the study area. The habitat variable et al. 1993; Boyce et al. 2002). The continuous Boyce Index with the highest marginality value has the most influence in value was calculated as a measure of the increase in the mean determining the species’ distribution (Hirzel et al. 2002). The predicted/expected (P/E) ratio as habitat suitability increases, second index, specialisation, shows the extent to which the using a Spearman rank correlation coefficient (Boyce et al. use of habitat variables by the species is narrow compared 2002). Results can vary from −1 to 1, with absolute values close with its overall distribution in the study area. Specialisation to 1 indicating that the model is not different from a random is calculated as the ratio of the standard deviation of the study model. Positive values indicate that the model correctly predicts area distribution to that of the species’ distribution. In ENFA, presences based on habitat suitability values, while negative the first axis accounts for all marginality of the species and Boyce Index values indicate the model has poor predictive some of the specialisation. The second and subsequent axes power (Hirzel et al. 2006b). are then extracted orthogonally to explain the remaining specialisation of the species (Hirzel et al. 2002). Most of the information explained by marginality and specialisation is Results contained in the first few axes. To compare the ecological niche of breeders and non- Between 15 March and 30 May 2006 we recorded 506 foraging breeders we ran two ENFA models, one using kākāpō foraging locations for the 18 adult female kākāpō in this study, using locations from 10 breeding females and the other from eight 482 triangulations and 24 sightings. The number of locations non-breeding females. Foraging locations were buffered to a collected per individual ranged from 17 to 34, with a mean 20-m radius to account for triangulation error and converted to (±SD) of 28.1 ± 4.5. The mean error associated with our a 50-m grid, using ArcGIS 9.1 (Esri Inc. Redlands, CA, 2005). triangulation technique was calculated to be 19.3 ± 12.2 m. We were limited to using a grid resolution of 50 m as this was There was no relationship between the number of location points the margin of error estimated in the vegetation map. used to estimate a home range and its size for all individuals Nine GIS-based raster layers were used to represent (R2 = 0.05; P = 0.38). potential kākāpō habitat on Codfish Island at a spatial resolution Foraging home ranges varied greatly in size between of 50 m. Elevation, slope and aspect were derived from a individuals for each of the home range methods (MCP: 3.1 – digital elevation model, using spatial analyst functions in 33 ha; 95% kernels: 3.5 – 26.5; 75% kernels: 2.4 – 16.6 ha; ArcGIS 9.1. Kākāpō may prefer short vegetation (Atkinson & 50% kernels: 1.4 – 9.7 ha) (Fig. 2), a finding consistent with Merton 2006; Butler 2006), so vegetation types with the same previous research on kākāpō home ranges (Moorhouse 1985; estimated canopy heights were merged to create three layers Trinder 1998; Farrimond et al. 2006; Walsh et al. 2006). MCP with canopy heights of approximately 20 m, 15 m and 5 m or and 95% kernel methods appeared to overestimate home range less (Table 2). The high-, moderate- or low-abundance classes size, a criticism commonly cited for both methods (White & of mature rimu trees were used as the final three habitat layers Garrott 1990; Seaman et al. 1999; Börger et al. 2006), while in the ENFA as vegetation selection analysis had indicated the the 50% kernel provided only an estimate of core foraging importance of rimu forest to breeding females. area. The 75% kernel home ranges closely resembled the areas ENFA requires quantitative rather than categorical data, so the canopy height and rimu-abundance layers were converted covered by foraging locations and excluded major outliers, so to binary raster grids, where 1 or 0 represented the presence were considered the most accurate representation of kākāpō or absence of the habitat variable, respectively. The focal foraging areas and these borders were used in later analysis statistics function of ArcGIS 9.1 was used to convert the (Fig. 3). layers from binary to continuous data, using a circular area Comparisons of foraging home ranges between breeders the mean size of the 75% kernel home range (7.11 ha with a (females that nested in 2005) and non-breeders (females that radius of 85 m). did not nest in 2005) showed some significant results. Breeders The first few factors resulting from the ENFA ofthe had foraging ranges on average twice the size of those of non- breeders’ model were used to predict suitable breeding breeders, with these differences being statistically significant habitat for female kākāpō on Codfish Island in a year of low for three of the four home range estimation techniques (Fig. food supply. A habitat suitability map was constructed using 2). Using the MCP method breeders had a significantly larger the distance geometric-mean algorithm as recommended by mean (±SD) home range size of 13.5 ± 8.2 ha, compared with Hirzel and Arlettaz (2003). The explained information was the mean non-breeders’ foraging range of 7.0 ± 3.6 ha (t12.778 maximised by determining the number of factors to include, = 2.247, P = 0.043). The 95% kernel method also estimated using a comparison of the factors’ eigenvalues based on the mean foraging range of breeders to be significantly larger MacArthur’s broken-stick distribution (Hirzel et al. 2002). at 15.1 ± 7.5 ha compared with the 8.0 ± 4.2 ha mean foraging Habitat suitability values ranging from 0 to 1 were computed for range estimated for non-breeders (t16 = 2.238, P = 0.030). The each cell by delineating envelopes around various proportions 75% kernel method showed that the breeders’ mean foraging of kākāpō locations and by counting the proportion of kākāpō range of 8.9 ± 4.5 ha was larger than the non-breeders’ locations they encompassed (Hirzel et al. 2006b). The map foraging range of 4.9 ± 2.5 ha, but these differences were not was reclassified into the following four habitat suitability quite statistically significant (W = 61.5, P = 0.062). The 50% classes: unsuitable (< 0.25), marginal (0.26 – 0.50), suitable kernel method estimating core foraging area showed that the (0.51 – 0.75) and optimal habitat (> 0.76). breeders’ mean core area was significantly larger at 5.0 ± 2.4 Whitehead et al.: Breeding variation in female kakapo 69

Figure 2. Comparison of mean (± 1 SE) foraging home range sizes for breeders (dark bars) and non-breeders (light bars) for the four home range estimation techniques used. Differences between breeders and non-breeders were statistically significant for all techniques except 75% kernels that were almost significant.

Figure 3. Distribution of forest with varying abundance of mature rimu trees on Codfish Island and the location of 75% kernel foraging ranges for females that bred during 2005 (breeders) and those that did not breed (non-breeders). 70 New Zealand Journal of Ecology, Vol. 36, No. 1, 2012

ha than the non-breeders’ mean core area of 2.2 ± 1.2 ha (W Multivariate habitat analysis = 67.5, P = 0.016; Fig. 2). The main difference between the habitat of breeders and non- Tenure was correlated with productivity as breeding breeders as detected by the ENFA models was the abundance females had on average been resident on Codfish Island for of tall, rimu-dominated forest (Table 5). Results of marginality longer than non-breeders (W = 68.5, P = 0.009). However, for the breeders’ model showed that the locations of breeding tenure was not correlated with home range size (z = 0.927, females were strongly biased towards forest containing a high P = 0.354). (0.472) or moderate (0.204) abundance of mature rimu trees Female kākāpō did not use vegetation at random on Codfish and biased against forest containing no mature rimu trees 2 2 Island (breeders: χ 4 = 77, P < 0.01; non-breeders: χ 4 = 34.8, (−0.433). In contrast, non-breeders’ locations were weakly P < 0.01) but appeared to select for certain types. Vegetation correlated with forest containing no mature rimu trees (0.152), use differed between breeders and non-breeders (Table 3). while they had no significant correlation with high- (0.012) or Although only 20% of the island contained rimu–miro forest, moderate-abundance (−0.029) rimu forest (Table 5). Breeders this vegetation type was heavily used by breeders, making tended to occupy tall forest with a maximum canopy height of up 39% of their foraging ranges compared with 25% of up to 20 m (0.233). Non-breeders occurred more frequently non-breeders’ ranges. Breeders were not recorded in coastal in short vegetation types with a maximum canopy height less daisy – pakahi scrub yet this vegetation made up 25% of non- than 5 m (0.318) and less often in vegetation types with a breeders’ foraging ranges. Both breeders and non-breeders used maximum canopy height of up to 15 m (−0.282). rātā–podocarp forest more than would have been expected Overall the most important factor in determining kākāpō assuming random habitat selection (Table 3). locations was elevation as it produced the largest absolute Vegetation with a high, moderate or low abundance of marginality values in both the breeders (0.627) and non- mature rimu trees each occupied around one-third of the island’s breeders (0.710) models (Table 5). Kākāpō distributions area (Table 4). Forest with a low abundance of rimu mostly were negatively correlated with slope, with both breeders occurred in coastal regions, while high-abundance rimu forest (−0.313) and non-breeders (−0.515) occurring in areas of occupied central, higher elevation areas (Fig. 3). There was a the island that were flatter than the mean available. Aspect significant difference in the quantity of high-abundance rimu did not appear to influence the location of breeders (−0.042) forest in foraging ranges of breeders and non-breeders (W = or non-breeders (−0.154) (Table 5) significantly, with both 65.5, P = 0.026), with high-abundance rimu forest occurring marginality coefficients showing only a small difference from in 60% of breeders’ foraging ranges compared with only the mean aspect available on the island. 31% of non-breeders’ (Table 4). Low-abundance rimu forest Evaluation of the ENFA models resulted in a continuous was a lot more common in non-breeders’ ranges, accounting Boyce Index of 0.25 ± 0.46 for the breeders’ model and 0.25 for 45% of vegetation, compared with just 9% for breeders, ± 0.65 for the non-breeders’ model. The positive values although these differences were not statistically significant indicate that, on the whole, the models correctly predicted (W = 21, P = 0.097). habitat suitability (Hirzel et al. 2006b) but the relatively low

Table 3. Proportions of aggregated vegetation classes (Table 2) on Codfish Island and within the foraging home ranges (75% kernel) of females that bred (breeders) and did not breed (non-breeders) in the low-rimu-mast year of 2005. ______Aggregated vegetation classes Proportion of each vegetation class on Codfish Island, and inside foraging ranges of breeders and non-breeders

Island Breeders Non-breeders ______Other 0.32 0.12 0.03 Coastal daisy – pakahi scrub 0.23 0.00 0.25 Rimu–miro 0.20 0.39 0.25 Rātā–podocarp 0.14 0.33 0.40 Mixed podocarp stunted 0.12 0.16 0.07 ______

Table 4. Comparisons of the relative abundance of mature rimu trees in the vegetation on Codfish Island and inside the foraging ranges (75% kernel) of females that bred (breeders) and did not breed (non-breeders) in the 2005 breeding season. (The proportions in bold highlight the distinct differences in high and low rimu abundances between breeders’ and non- breeders’ foraging ranges). ______Relative abundance of mature Proportion of rimu abundance on Codfish Island, and inside foraging ranges of breeders rimu trees in vegetation and non-breeders

Island Breeders Non-breeders ______High rimu abundance 0.37 0.60 0.31 Moderate rimu abundance 0.32 0.31 0.24 Low rimu abundance 0.31 0.09 0.45 ______Whitehead et al.: Breeding variation in female kakapo 71

Table 5. Marginality (in rank order) and specialisation coefficients (SC) are shown for the nine habitat variables included in the ENFA breeders and non-breeders models for the first six ecological factors. ______Marginality SC1 SC2 SC3 SC4 SC5 ______Breeders (27%) (33%) (15%) (7%) (5%) (5%) Elevation 0.627 −0.08 −0.459 0.147 0.331 0.034 Frequency of high-abundance rimu forest 0.472 0.159 0.551 −0.576 −0.364 0.346 Frequency of no rimu forest −0.433 −0.193 0.153 −0.439 −0.033 0.045 Slope −0.313 0.083 −0.321 0.127 −0.338 0.182 Frequency of up to 20-m canopy 0.233 0.01 −0.279 0.486 −0.276 −0.588 Frequency of moderate rimu forest 0.204 −0.52 0.256 −0.382 −0.393 0.158 Aspect −0.042 −0.003 −0.035 0.208 −0.004 −0.079 Frequency of up to 5-m canopy 0.027 0.808 −0.125 0.073 −0.577 −0.521 Frequency of up to 15-m canopy −0.016 0.003 −0.445 −0.084 −0.281 −0.443 ______Non-breeders (25%) (31%) (14%) (9%) (7%) (5%) Elevation 0.710 −0.286 0.289 0.248 −0.115 0.113 Slope −0.515 0.08 0.205 0.247 −0.209 0.102 Frequency of up to 5-m canopy 0.318 0.431 −0.41 0.379 −0.527 −0.612 Frequency of up to 15-m canopy −0.282 −0.343 −0.065 0.288 −0.39 −0.542 Aspect −0.154 0.035 0.087 0.303 0.158 −0.108 Frequency of no rimu forest 0.152 −0.082 −0.1 −0.023 0.284 −0.188 Frequency of up to 20-m canopy −0.051 −0.371 −0.713 0.739 −0.323 −0.455 Frequency of moderate rimu forest −0.029 −0.026 0.027 −0.045 0.326 −0.146 Frequency of high-abundance rimu forest 0.012 0.68 0.418 −0.101 0.444 −0.182 ______ENFA = Ecological Niche Factor Analysis (Hirzel et al. 2002) Notes: Habitat variables are sorted by decreasing absolute value of coefficients on the marginality factor. Positive values on this factor mean that adult female kākāpō prefer locations with higher values on the corresponding habitat variable than the mean location on the island. Signs of coefficients have no meaning on the specialisation factors. The amount of specialisation accounted for is given in brackets in each column for breeders and non-breeders.

magnitude of the index values combined with large standard likely resulted in an underestimation of the actual area used errors, particularly for the non-breeders’ model, indicates a by kākāpō for foraging (Trinder 1998). low degree of model robustness (Sattler et al. 2007). Importantly, our results show that home range sizes of Six significant factors of the breeder’s ENFA model were breeding females are almost twice that of non-breeding females, retained for computing a habitat suitability map predicting and are largely non-overlapping (Fig. 4), regardless of the optimal breeding habitat for female kākāpō in low-rimu-mast method employed for home range estimation. This suggests years (Fig. 4). Together the six factors explained 96% of the that there is a clear reproductive benefit to kākāpō with larger information contained in all variables (100% of the marginality home ranges and that this habitat is generally not shared, at least and 92% of the specialisation). Optimal breeding habitats were not with other females. Our results also indicate that females mostly located in the central regions of the island at higher with a longer tenure on the island are more likely to breed, elevations (Fig. 4). Some large areas of optimal breeding although this relationship is not mediated by home range size, habitat were not occupied by adult female kākāpō, possibly but perhaps by prior opportunity to occupy prime habitats. because they were inhabited by any of the 33 other kākāpō on Thus, the implication is that individual kākāpō females on the island including dominant adult males. Females that did Codfish Island have effectively partitioned available habitat, not breed in 2005 were mostly located in habitat classified as both in space and time, based on the availability and quality marginal or unsuitable for breeding (Fig. 4). of resources for successful reproduction. Both the vegetation selection and ENFA multivariate analyses suggest that females were limited in their ability to Discussion breed if their access to mature rimu forest was limited. The importance of rimu fruit abundance in triggering kākāpō to This study is the first to characterise home ranges based on breed on Codfish Island (Harper et al. 2006) and in increasing the night-time foraging activities of female kākāpō on Codfish breeding attempts (Elliott et al. 2006) at a population level Island, as well as to make comparisons of foraging home range is well recognised. Our results go a step further by showing sizes between breeding and non-breeding female kākāpō. that at an individual level the ability of each female kakapo Previous reports of kākāpō home range sizes on various to breed in a low food supply year is limited by their access offshore islands, including Codfish Island, have been mainly to food resources. Breeding females appear to select for based on day-time roosting locations (e.g. Best & Powlesland certain vegetation types, particularly those containing rimu, 1985; Moorhouse 1985; Farrimond et al. 2006), which have at the home range scale. For individual females on Codfish 72 New Zealand Journal of Ecology, Vol. 36, No. 1, 2012

Figure 4. Relative suitability of habitat for adult female kākāpō to breed on Codfish Island in low-rimu-mast years (as predicted by the ENFA breeders’ model), compared with actual foraging ranges (75% kernel) for females that bred (breeders, n = 10) and did not breed (non-breeders, n = 8) in 2005. Locations recorded during this study are also shown for the three adult females not radio-tracked, 20 adult males and 13 juvenile kākāpō also present.

Island, rimu availability is influenced by both the size and of the vegetation in describing high-quality breeding habitat. habitat composition of the home range. In 2005, only females Locations of breeding females were strongly positively that had home ranges with a moderate-to-high prevalence of correlated with the presence of high-abundance rimu vegetation rimu were able to breed in this relatively low rimu mast year. types and strongly negatively correlated with vegetation types In high-rimu-mast years, such as 2002, the abundance of with a low abundance of rimu forest. Elevation was given the fruit apparently overwhelms the importance of home range highest importance value in models describing both breeding characteristics, as all females were able to obtain sufficient and non-breeding female kākāpō habitat, indicating that, in fruit from fewer trees, resulting in successful nesting attempts general, individuals prefer to inhabit higher elevation portions by nearly all females (Table 1). of the island. Indeed, past descriptions of kākāpō distributions The importance of rimu forest in relation to kākāpō and resource use have shown that kākāpō may prefer higher breeding was further supported by the ecological niche factor elevation ecotonal areas, such as those found near treelines analysis that assessed the relative importance of a range of and slips and where the birds can more easily access a variety habitat variables to kākāpō breeding habitat. With the exception of food sources (Atkinson & Merton 2006; Butler 2006). of elevation, the relative abundance (as measured by the Despite the generally low robustness reported for the ENFA relative degree of prevalence) of rimu forest had the highest models for both breeders and non-breeders, variables chosen marginality score, and was therefore considered more important for inclusion in the models were statistically important, and than other variables such as slope, aspect or the canopy height habitat suitability maps produced from the models (Fig. 4) Whitehead et al.: Breeding variation in female kakapo 73

provided insight into the overall spatial distribution of kākāpō available. Further, translocation of any juvenile kākāpō or relative to the predicted distribution of habitat of differing adult males that are not currently required in the breeding quality. Such maps can be useful tools for conservation as population, to a non-breeding island, might allow breeding-age they may, for instance, enable managers to identify likely females to increase the sizes of their foraging home ranges on candidates for translocation to new islands. By overlaying Codfish Island and, thus, the quality of the breeding habitat foraging home ranges onto the breeders habitat suitability map to which they have access. (Fig. 4) we were able to show that females only needed a small Through this research we now know that there are factors area of optimal breeding habitat in their home range to enable limiting the ability of adult female kākāpō to breed in low- them to breed even in low-rimu-mast years, provided that the rimu-mast years on Codfish Island. Perhaps more importantly remainder of their foraging home ranges were mostly located we also know that these factors – home range size and access in moderately suitable breeding habitat. The overlays also to mature rimu forest – can be manipulated by conservation showed that non-breeding females mostly occupied unsuitable managers to give individual birds the best chance of being or marginal habitats despite there being large areas of optimal able to breed each breeding season. The principles applied breeding habitat not occupied by breeding-age females. in this management-focused research could easily be applied Locations collected for the 33 other kākāpō inhabiting the to other rare species to gain a better understanding of their island during the study period, and overlaid onto the habitat ecology and the importance of a range of variables in their suitability map (Fig. 4), showed that juveniles were largely breeding performance. inhabiting coastal areas of the island within marginal habitat; some adult males, on the other hand, were found in areas of optimal or suitable breeding habitat, suggesting that they may Acknowledgements compete with females for higher quality habitat. We suggest that further investigations into kākāpō We thank staff in the Department of Conservation’s Kakapo habitat use and resource selection would benefit from an Recovery Programme for their support of this project, in improved characterisation of the variability in kākāpō resource particular Ron Moorhouse, Graeme Elliott and Daryl Eason. composition, quantity, and quality at the scale relevant to Thank you to the numerous volunteers who assisted with data foraging kākāpō. A drawback of this study was the limited collection and those who reviewed and supported us with spatial resolution and compositional detail of the vegetation data this manuscript, with special thanks to Amy Whitehead. We used in our analyses, which limited our ability to investigate acknowledge the roles and contributions of each of the authors within-home-range resource selection. Aside from the overall of this paper as follows: J. Whitehead (main contributor of importance of rimu abundance, the exact form and mechanisms research, analysis and writing); B. Case (assistance with spatial of the available cues used by kākāpō to trigger nesting (Harper analysis and writing); L. Molles (assistance with fieldwork, et al. 2006), and the importance of other fine-scale resources statistics and writing); K-J. Wilson (assistance with project for nutrition (Elliot et al. 2001), remain unclear. As such, data development, fieldwork and writing). Financial support for on plant composition and abundance at different spatial scales, this project was received from Lincoln University, a Todd in combination with more direct tracking of movements (e.g. Foundation Award for Excellence, a Freemasons Postgraduate with GPS receivers) and dietary habits of kākāpō of varying Scholarship, a Gordon Williams’ Postgraduate Scholarship age throughout the year, would help shed light on some of and a Lady Isaac Postgraduate Scholarship. these issues.

References Conclusions Atkinson IAE, Merton DV 2006. Habitat and diet of kakapo The results of this research support two general hypotheses (Strigops habroptilus) in the Esperance Valley, Fiordland, posed at the outset of the study. Specifically we found that: New Zealand. Notornis 53: 37–54. (1) female kākāpō that bred during the low-rimu-mast year Best HA, Powlesland RG 1985. Kakapo. Dunedin, John of 2005 had larger foraging home ranges relative to females McIndoe and , New Zealand Wildlife Service. that did not breed that year; (2) home ranges of those that 30 p. bred were situated in areas of higher quality breeding habitat, Börger L, Franconi N, de Michele G, Gantz A, Meschi F, Manica characterised in particular by a higher prevalence of mature A, Lovari S, Coulson T, 2006. Effects of sampling regime rimu forest. We therefore conclude that female kākāpō are on the mean and variance of home range size estimates. partitioning available space and resources on Codfish Island Journal of Ecology 75: 1393–1405. in ways that increase the likelihood of breeding. Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA The trend of reduced productivity of adult female kākāpō 2002. Evaluating resource selection functions. Ecological on Codfish Island in low-rimu-mast years is most likely caused Modelling 157: 281–300. by some individuals having insufficient critical resources to breed, a finding that will be valuable to managers as they try to Brotons L, Thuiller W, Araujo MB, Hirzel AH 2004. Presence– increase the productivity of this critically endangered species. absence versus presence-only modelling methods for As the size of a home range is likely to be dependent on the predicting bird habitat suitability. Ecography 27: 437–448. number of potential competitors, it could be expected that as Butler DJ 2006. The habitat, food and feeding ecology of the Codfish Island kākāpō population grows, home ranges will kakapo in Fiordland: a synopsis from the unpublished likely reduce in size. Females in mostly unsuitable or marginal MSc thesis of Richard Gray. Notornis 53: 55–79. breeding habitat are unlikely to breed in low-rimu-mast years Clout MN 2006. A celebration of kakapo: progress in and so could be considered as candidates for translocation to conservation of an enigmatic parrot. Notornis 53: 1–2. another island where more suitable breeding habitat may be Cockrem JF 2006. The timing of breeding in the kakapo 74 New Zealand Journal of Ecology, Vol. 36, No. 1, 2012

(Strigops habroptilus). Notornis 53: 153–159. Merton DV, Morris RB, Atkinson IAE 1984. Lek behaviour in Cockrem JF, Rounce JR 1995. Non-invasive assessment of a parrot: the kakapo Strigops habroptilus of New Zealand. the annual gonadal cycle in free-living kakapo (Strigops Ibis 126: 277–283. habroptilus) using fecal steroid measurements. Auk 112: Meurk CD, Wilson HD 1989. Stewart Island. Biological 253–257. Survey of Reserves series 18. Wellington, Department Eason DK, Elliott GP, Merton DV, Jansen PW, Harper GA, of Conservation. 162 p. Moorhouse RJ 2006. Breeding biology of kakapo (Strigops Millspaugh JJ, Marzluff JM eds 2001. Radio tracking and habroptilus) on offshore island sanctuaries, 1990–2002. animal populations. San Diego, Academic Press. 474 p. Notornis 53: 27–36. Moorhouse RJ 1985. Ecology of kakapo (Strigops habroptilus Elliott GP, Merton DV, Jansen PW 2001. Intensive management Gray, 1845) liberated on (Hauturu). of a critically endangered species: the kakapo. Biological Unpublished MSc thesis, University of , Conservation 99: 121–133. New Zealand. 151 p. Elliott GP, Eason DK, Jansen PW, Merton DV, Harper GA, Moorhouse RJ 2009. Kakapo Programme Annual Report 1 July Moorhouse RJ 2006. Productivity of kakapo (Strigops 2008 – 30 June 2009. Southern Islands Area, Department habroptilus) on offshore island refuges. Notornis 53: of Conservation. 138–142. Moorhouse RJ, Powlesland, RG 1991. Aspects of the ecology Farrimond M, Clout MN, Elliott GP 2006. Home range size of of kakapo (Strigops habroptilus) liberated on Little Barrier kakapo (Strigops habroptilus) on Codfish Island. Notornis Island (Hauturu). New Zealand. Biological Conservation 53: 150–152. 56: 349–365. Fidler AE, Lawrence SB, McNatty KP 2008. An hypothesis to Powlesland RG, Lloyd BD, Best HA, Merton DV 1992. explain the linkage between kakapo (Strigops habroptilus) Breeding biology of the kakapo Strigops habroptilus on breeding and the mast fruiting of their food trees. Wildlife Stewart Island, New Zealand. Ibis 134: 361–373. Research 35: 1–7. R Development Core Team 2008. R: A language and Harper GA, Elliott GP, Eason DK, Moorhouse RJ 2006. environment for statistical computing. Vienna, Austria, What triggers nesting of kakapo (Strigops habroptilus)? R Foundation for Statistical Computing. http://www. Notornis 53: 160–163. R-project.org. Harris S, Cresswell WJ, Forde PG, Trewhella WJ, Woollard T, Sattler T, Bontadina F, Hirzel AH, Arlettaz R 2007. Ecological Wray S 1990. Home-range analysis using radio-tracking niche modelling of two cryptic bat species calls for a data – a review of problems and techniques particularly reassessment of their conservation status. Journal of as applied to the study of mammals. Mammal Review Applied Ecology 44: 1188 – 1199. 20: 97–123. Seaman DE, Powell RA 1996. An evaluation of the accuracy Hirzel AH, Arlettaz R 2003. Modelling habitat suitability of kernel density estimators for home-range analysis. for complex species distributions by environmental- Ecology 77: 2075–2085. distance geometric mean. Environmental Management Seaman DE, Millspaugh JJ, Kernohan BJ, Brundige GC, 32: 614–623. Raedeke KJ, Gitzen RA 1999. Effects of sample size Hirzel AH, Hausser J, Chessel D, Perrin N 2002. Ecological- on kernel home range estimates. Journal of Wildlife niche factor analysis: How to compute habitat-suitability Management 63: 739–747. maps without absence data? Ecology 83: 2027–2036. Springer JT 1979. Some sources of bias and sampling error Hirzel AH, Hausser J, Perrin N 2006a. Biomapper 3.2. A GIS- in radio triangulation. Journal of Wildlife Management toolkit to model ecological niche and habitat suitability. 43: 926–935 Switzerland, Division of Conservation Biology, University Trinder M 1998. A comparison of day and night home-range of Bern. URL: http://unil.ch/biomapper size in the New Zealand kakapo Strigops habroptilus. Hirzel AH, Le Lay G, Helfer V, Randin C, Guisan A 2006b. Unpublished MSc thesis, University of Reading, UK. Evaluating the ability of habitat suitability models to Walsh JE 2002. Seasonal changes in home range size and predict species presences. Ecological Modelling 199: habitat selection by kakapo (Strigops habroptilus) on 142–152. Maud Island. Unpublished MSc thesis, Lincoln University, Kenward RE, South AB, Walls SS 2003. Ranges6 v1.2: For New Zealand. 105 p. the analysis of tracking and location data. Wareham, UK, Walsh J[E], Wilson K-J, Elliott GP 2006. Seasonal changes in Anatrack. home range size and habitat selection by kakapo (Strigops Kernohan BJ, Gitzen RA, Millspaugh JJ 2001. Analysis of habroptilus) on Maud Island. Notornis 53: 143–149. animal space use and movements In: Millspaugh JJ, White GC, Garrott RA 1990. Analysis of wildlife radio-tracking Marzluff JM eds Radio tracking and animal populations. data. San Diego, CA, Academic Press. 383 p. San Diego, Academic Press. Pp. 125–166. Whitehead JK 2007. Breeding success of adult female kakapo Lignon JD 1974. Green cones of the piñon pine stimulate late (Strigops habroptilus) on Codfish Island (Whenua Hou): summer breeding in the piñon jay. Nature 250: 80–82. correlations with foraging home ranges and habitat Manly BFJ, McDonald LL, Thomas DL 1993. Resource selection. Unpublished MSc thesis, Lincoln University, selection by : statistical design and analysis of field New Zealand. 134 p. studies. 1st edn. London, Chapman and Hall. 177 p. Worton BJ 1989. Kernel methods for estimating the Mech LD 1983. Handbook of animal radio-tracking. utilization distribution in home-range studies. Ecology Minneapolis, MN, University of Minnesota Press. 70: 164–168. 107 p.

Editorial Board member: Phil Lester Received 3 June 2010; accepted 7 April 2011